This month's VR and augmented reality category includes 6 filed patent applications from Sony (2), Apple (1), Beijing Zitiao Network Technology (1), Meta (1), and Niantic (1).
The patents cover a range of techniques spanning avatar design, AR content localization, and rendering efficiency, with Meta describing a system for deriving consistent design-language constraints across platforms through parameterized avatar generation, and Niantic filing an approach that localizes AR content in new environments using a single scene-agnostic coordinate model paired with lightweight per-scene map codes. Sony's 2 applications address pulling 3D game characters into a player's physical space via AR headset and concealing tracking-mode transitions by timing switches to a user's natural blink or saccade. Apple and Beijing Zitiao Network Technology round out the category with methods focused on causal reprojection, gaze prediction, and foveated rendering to reduce power consumption and stutter in wearable XR headsets.
Meta received 1 patent covering how avatar design languages can be turned into measurable generation constraints. Rather than requiring artists to manually define every proportion or acceptable range for a character's features, the system randomly generates test avatars, measures anatomical landmarks like eyes, ears, nose, and lips, and retains only those that conform to the intended style. The constraints derived from that process can then be applied to parameterize new avatars consistently across social, XR, and game platforms.
Sony received 2 patents in this category, both addressing how virtual content intersects with a user's physical or perceptual reality. The first describes a method for intercepting 3D character models from a running game and re-rendering them in the user's real-world environment through an AR headset, all without any modification to the underlying game program itself. The second addresses a more subtle problem: when a VR headset loses camera or IMU tracking confidence, switching modes abruptly can produce a jarring visual jump, so the application gates that transition on a detected eye event, either a blink or a saccade, timing the switch to a moment when the user is least likely to perceive the discontinuity.
Niantic filed 1 patent describing a localization approach that avoids the need to build a 3D model of a location or train a dedicated model for each new scene. The system separates what it knows into 2 parts: a single scene-agnostic coordinate model stored in the network's parameters, and a compact per-scene map code trained in alternating phases alongside it. Adding a new location requires only generating a new map code rather than constructing geometry or retraining the model from scratch.
Apple filed 1 patent describing a deep learning method that addresses both frame generation and foveated rendering on wearable headsets. A causal reprojection network uses body position parameters and scene geometry to predict additional frames temporally, rather than relying solely on prior image frames, which allows the headset to produce smoother output without a proportional increase in rendering load. The same body position parameters also feed a gaze prediction step that estimates where the user will be looking, letting the system render the gaze region and the periphery at separate resolutions.
Beijing Zitiao Network Technology received 1 patent for a rendering approach that splits a frame into 2 temporal streams based on where the user is looking. The gaze region refreshes at a rate at least 2 times higher than the full frame, while the surrounding area updates more slowly, reducing the total rendering burden for XR devices. The core idea is a frame rate split rather than a resolution or quality split, trading update frequency in the periphery for lower power consumption and reduced stutter.
All data sourced from USPTO patent filings. Google Patents may take several weeks to index recent publications. If a link is unavailable, search for the patent number at USPTO Patent Public Search.